Smishing Text Analysis Web Application

This Django Web Application is a wrapper for a trained voting classifier model that combines the results of 6 different classifiers through soft voting. The voting classifier has been trained on a dataset, which combines a public dataset and 240 smishing texts that have been generated using Chat-GPT 3.5.

Installation

It is recommended to run this project on a linux based environment.

Download the code, either as a zip file or through git clone:

git clone https://github.com/Denno20/Smishing-detector-web-app.git

Once downloaded, head into the root of the directory where manage.py is located.

It is best practice use a virtual environment to isolate package installations from other projects. To create a virtual environment, go to the project root directory and run the following command. This will create a new virtual environment in a local folder named .venv. After this has installed, activate the virtual environment.

python3 -m venv .venv
source .venv/bin/activate

Use the package manager pip to install the packages in the requirements.txt file. This may take a while so please be patient.

pip3 install -r requirements.txt

GUI Usage

To start the web server, run the following command in the project root directory: ~/SmishingSMSPredictor and navigate to http://127.0.0.1:8000/ in your web browser.

python3 manage.py runserver

Command Line Usage

The application creates an instance of the combinedClassifier.py class, which can be used in the command line like a library. Using this library, an instance of the CombinedClassifier model can be instantiated which can be trained, saved and loaded for later use.

#Import libraries
import pandas as pd
from combinedClassifier import CombinedClassifier

#Create classifier
model = CombinedClassifier()
model.load_dataset("./Datasets/combined_dataset.csv")

#The option to save the model appears in this function
#WARNING: Saving the model will OVERWRITE the current model joblib files saved in memory
model.train()

The model saves 2 joblib files, one for the model and one for the text vectoriser. These files need to be loaded into the model before it can make predictions:

model = CombinedClassifier()
model.load_model("voting_classifier.joblib", "vectorizer.joblib")

The model can then make predictions on unseen data using the following code. The unseen data must be in a python list and then converted to a Series using pandas.

#Example data
unseen = ["You have won $5000. The prize needs to be claimed ASAP. Please reply with your bank information so we can deposit the money into your account.", "Hey, see you soon!"]
predicted = model.predict(pd.Series(unseen))
print(predicted)

Dataset

The model can be trained on any smishing dataset using a csv file of the following structure:

Label	Text
smish	SMS GOES HERE
ham	SMS GOES HERE

Data must be labelled as either "smish" or "ham", and there must be two columns with the header "Label" and "Text".

Resources used for this application

• Machine learning tutorial.
• Machine learning voting classifier.
• Hypothesis testing code.
• Training Dataset.
• Scikit-learn.